10 tickets across 6 weeks. Two engineers, parallel tracks, zero idle time.
Architecture, VPC/IAM/security, core business logic (processor, ETL, comparison engine), SAP coordination, code review, cutover. ~40% of total work.
CDK stacks (S3, DynamoDB, ECS, CF), CI/CD, Lambda replication, React frontend, Docker, validation scripts, monitoring, tests. ~60% of total work.
| Ticket | Tech Lead | Full-Stack | Week |
|---|---|---|---|
| T1 Infra + CI/CD | 30% | 70% | 1 |
| T2 Storage Layer | 30% | 70% | 1 |
| T3 Ingestion Lambdas | 35% | 65% | 2 |
| T4 ETL Processing | 70% | 30% | 3 |
| T5 Shadow Validation | 20% | 80% | 3 |
| T6 Backend → ECS | 40% | 60% | 4 |
| T7 Frontend → CF | 15% | 85% | 5 |
| T8 Auth Fix | 65% | 35% | 4 |
| T9 Auto-Gen + Compare | 70% | 30% | 5 |
| T10 Monitor + Cutover | 35% | 65% | 6 |
Tiny data volume (~73 products, ~67 packaging, ~21 HTS). Lambda + pandas is 35x cheaper than Glue.
Lowest risk. Move users.json to DynamoDB. Defer Cognito/Entra ID until M365 decision is final.
| Risk | Mitigation |
|---|---|
| SAP team delays on 6 APIs | Build against mocks from day 1. Swap URLs when APIs ship. |
| 6 weeks is tight | Parallel tracks every week. No idle time. Daily standups. |
| Cross-account S3 complexity | Agreed Mar 26 meeting. Test in Week 1. |
| Auth migration breaks things | T8 isolated in Week 4. Test in staging first. |
| Processor output differs | Shadow mode (T5 Week 3) catches before cutover. |
| Scope creep | Document Distribution Pipeline is OUT. No exceptions. |